Table of Contents
Support for multiple JNI libraries
Firebird management interfaces
Incompatible changes since Jaybird 2.0
Using Type 2 and Embedded Server driver
Configuring Type 2 JDBC driver
Configuring Embedded Server JDBC driver
JDBC 3.0 deviations and unimplemented features
Using java.sql.ParameterMetaData with Callable Statements
Using ResultSet.getCharacterStream with BLOB fields
Heuristic transaction completion support
Connection pooling with Jaybird 2.0
org.firebirdsql.pool.FBConnectionPoolDataSource
org.firebirdsql.pool.FBWrappingDataSource
Runtime object allocation and deallocation hints
Jaybird is JCA/JDBC driver suite to connect to Firebird database server. Historically Borland opened sources of type 3 JDBC driver called InterClient. However due to some inherent limitations of Firebird client library it was decided that type 3 driver is a dead end, and Firebird team developed pure Java implementation of wire protocol. This implementation became basis for Jaybird, pure Java driver for Firebird relational database.
This driver is based on both the new JCA standard
for application server connections to enterprise information systems
and the well known JDBC standard. The JCA standard specifies an
architecture in which an application server can cooperate with a
driver so that the application server manages transactions, security,
and resource pooling, and the driver supplies only the connection
functionality. While similar to the JDBC 2 XADataSource
idea, the JCA specification is considerably clearer on the division
of responsibility between the application server and driver.
Jaybird supports Firebird 1.0.x, Firebird 1.5 and upcoming Firebird 2.0 SuperServer and Classic independently of the platform on which server runs when type 4 JDBC driver is used. Type 2 and embedded server JDBC drivers require JNI library, precompiled binaries for Win32 and Linux platforms are shipped in the default installation, other platforms require porting/building JNI library for that platform.
Driver no longer supports InterBase servers due to a Firebird-specific changes in database attachment parameters that are send to the server.
Driver supports following specifications:
JDBC 3.0 |
Driver passed complete JDBC compatibility test suite, though some features are not implemented. It is not officially JDBC compliant, because of high certification costs. |
JDBC 2.0 Standard Extensions |
Jaybird provides implementation of following
interfaces from
|
JCA 1.0 |
Jaybird provides implementation of
|
JTA 1.0.1 |
Driver provides implementation of
|
JMX 1.2 |
Jaybird provides MBean to manage Firebird server and installed databases via JMX agent. |
Jaybird 2.1 introduces following new features:
Events is one of the unique features in the Firebird RDBMS and allows asynchronous notification of the applications about named events that happen in the database. The information about how to use this feature can found in free IB 6.0 documentation set as well as in The Firebird Book by Helen Borrie.
The interfaces and classes for the event support
can be found in org.firebirdsql.event
package, which includes:
EventManager
interface to register for the synchronous and asynchronous
notification about the events in the database;
EventListener
interface which has to be implemented by the application that wants
to participate in the asynchronous notification;
DatabaseEvent
interface which represents the object that will be passed to the
EventListener
notification method;
Implementation of the above interfaces:
FBEventManager
and FBDatabaseEvent
.
Please note, that each instance of
FBEventManager
will open a new socket
connection to the Firebird server on the port specified by the
Firebird.
Similar to other JDBC extensions in Jaybird, the interfaces are released under the modified BSD license, the implementation of the code is released under LGPL license.
Until this release only one client library could
be loaded in the single JVM. That could be either embedded Firebird
library (fbembed.dll
/libfbembed.so
),
or Firebird client library (fbclient.dll
/libfbclient.so
)
or Fyracle client library. This could lead to possible problems, for
example, if embedded Firebird was used first, JDBC driver would
access database file directly instead of using the local IPC protocol
if only path to the database was specified. It was not possible to
change this without restarting the JVM.
Now Jaybird is able to correctly load arbitrary
number of shared libraries that implement the ISC API and forward the
requests correctly depending on the type of the driver being used.
Note, this feature requires new JNI library
(jaybird21.dll
/libjaybird21.so
)
Note, the JNI library
(jaybird21.dll
/libjaybird21.so
)
was changed in a incompatible manner between Jaybird 2.1 beta 1 and
Jaybird 2.1 RC1 releases. Please ensure to use the latest version.
New connection property that allows to create holdable result sets by default. This is needed as an workaround for the applications that do not follow JDBC specification in regard to the auto-commit mode.
Specifically, such applications open a result set and, while traversing it, execute other statements using the same connection. According to JDBC specification the result set has to be closed if another statement is executed using the same connection in auto-commit mode. Among others the OpenOffice Base users have problems with the restored JDBC 3.0 compatibility in Jaybird 2.0.
The new property is called:
defaultResultSetHoldable
as connection property for JDBC URL or for java.sql.DriverManager
class and no or empty value should be assigned to it; it has an
alias defaultHoldable
to simplify the
typing;
isc_dpb_result_set_holdable
as a DPB member;
FirebirdConnectionProperties
interface was with two methods extended (the
isDefaultResultSetHoldable()
and
setDefaultResultSetHoldable(boolean)
)
which is also reflected in the FBConnectionPoolDataSource
and FBWrappingDataSource
classes.
Note, the price for using this feature is that each holdable result set will be fully cached in memory. The memory occupied by it will be released when the statement that produced the result set is either closed or re-executed.
StatisticsManager
now
provides methods to fetch statistics for the specified tables only
(like the -t switch in gstat) and to provide also analysis of average
record and version length.
Jaybird 2.0 underwent significant internal changes that are not directly visible to the client code, and not necessarily introduced new features. However those changes significantly increase code modularity, stability and maintainability.
Significant refactoring of the
org.firebirdsql.gds.*
package.
Implementations of the interfaces defined in this package are now
dynamically loaded from the classpath and provide possibilities to
extend driver with custom code without changing the driver itself.
JCA code was significantly changed, bringing a long awaited stability when critical errors happen into that vital part of code. There is no more so-called “fatal” errors that can cause driver to recycle connections and transactions.
Refactoring of the transaction and object life-cycle management in the JDBC code. Main reason for this change was the need to remove result set caching in auto-commit mode and make it JDBC specification compliant, but it has also positive effect on the code stability and maintainability.
Unified connection property handling in JCA, driver manager and pooling code. This solves a longstanding issue with different configuration possibilities in different modes.
Refactorings to support JDK 5.0.
Jaybird provides now support for the
updatable result sets. Feature allows Java application to update
current record using the updateXXX
methods of java.sql.ResultSet
interface.
Updates are performed within the current transaction using a best row
identifier in WHERE
clause. This sets
the following limitation on the result set “updatability”:
the SELECT
references single table;
all columns not referenced in
SELECT
permit NULL
s
(otherwise INSERT
s will fail);
the SELECT
statement does not contain DISTINCT
predicate, aggregate functions, joined tables or stored procedures;
the SELECT
statement references all columns from the table primary key
definition or an RDB$DB_KEY
column.
Jaybird provides full support of the Firebird Services API that allows Java applications to perform various server management tasks:
database backup/restore on remote server; it is possible to performs metadata-only backups, switch the garbage collection during backup off, restore databases with no validity constraints or active indices, etc.
database maintenance, e.g. database shutdown, sweep, changing the forced writes settings, changing SQL dialect of the database, shadow management, etc.
retrieving database statistics including header page statistics, system table statistics, data page statistics and index statistics.
user management, including adding, modifying, and deleting user accounts.
Signatures of all org.firebirdsql.gds.*
classes have changed. This was done in preparation of the official
Firebird Java API. After the API release all signature changes will
be frozen and correctly versioned.
FirebirdConnection.getBlob()
method signature changed to return java.sql.Blob
,
previously it returned org.firebirdsql.jdbc.FirebirdBlob
.
Object that is returned can safely be casted to
org.firebirdsql.jdbc.FirebirdBlob
interface. The change is caused by the forthcoming JDBC 4.0
specification to avoid conflict in next versions.
Jaybird 1.x allowed to execute queries in
auto-commit mode, loop through the obtained result set and to
execute other statements through the same connection. This is no
longer possible in Jaybird without changing the result set
holdability. The behavior was changed in order to conform the JDBC
3.0 specification, chapter 10.1 “Transaction Boundaries and
Auto-commit”. Applications willing to use this feature have to
set the result set holdability to
ResultSet.HOLD_CURSORS_OVER_COMMIT
.
Please note, that this option assumes that result set type is
TYPE_SCROLL_INSENSITIVE
and is
completely cached on the client side.
Jaybird provides extensions to some
JDBC interfaces. JDBC extension interface classes are released under
modified BSD license, on “AS IS” and “do what you
want” basis, this should make linking to these classes safe
from the legal point of view. All classes belong to
org.firebirdsql.jdbc.*
package. Table
below shows all JDBC extensions present in Jaybird with a driver
version in which the extension was introduced.
JDBC extensions |
|||
---|---|---|---|
Interface |
Since |
Method name |
Description |
|
2.0 |
|
Create new instance of
|
|
|
|
Connect to the Firebird database using the specified connection properties. |
|
2.0 |
see JDBC connection properties section for more details. |
|
|
1.5 |
|
Create new BLOB in the database.
Later this BLOB can be passed as a parameter into
|
|
1.5 |
|
Get connection character encoding. |
|
2.0 |
|
Get the TPB parameters for the specified transaction isolation level. |
|
2.0 |
|
Create an empty transaction parameter buffer. |
|
2.0 |
|
Set TPB parameters for the specified transaction isolation level. The newly specified mapping is valid for the whole connection lifetime. |
|
2.0 |
|
Set TPB parameters for the specified transaction isolation level. The newly specified parameters are effective until the transaction isolation is changed. |
|
|
|
Get source code for the specified stored procedure name. |
|
|
|
Get source code for the specified trigger name. |
|
|
|
Get source code for the specified view name. |
|
1.5 |
|
Extension that allows to get more precise information about outcome of some statement. |
|
1.5 |
|
Check if this statement has open result set. Correctly works only when auto-commit is disabled. Check method documentation for details. |
|
1.5 |
|
Get current result set. Behaviour
of this method is similar to the behavior of the
|
|
1.5 |
|
Check if this statement is still valid. Statement might be invalidated when connection is automatically recycled between transactions due to some irrecoverable error. |
|
2.0 |
|
Get execution plan for the last executed statement. |
|
2.0 |
|
Get the execution plan of this prepared statement. |
|
2.0 |
|
Get the statement type of this prepared statement. |
|
1.5 |
|
Mark this callable statement as a call of the selectable procedure. By default callable statement uses "EXECUTE PROCEDURE" SQL statement to invoke stored procedures that return single row of output parameters or a result set. In former case it retrieves only the first row of the result set. |
|
2.0 |
|
Get execution plan for this result set. |
|
1.5 |
|
Method “detaches” a BLOB object from the underlying result set. Lifetime of “detached” BLOB is limited by the lifetime of the connection. |
|
1.5 |
|
Check if this BLOB is segmented. Seek operation is not defined for the segmented BLOBs. |
|
1.5 |
|
Opens an output stream at the specified position, allows modifying BLOB content. Due to server limitations only position 0 is supported. |
|
1.5 |
|
Get corresponding BLOB instance. |
|
1.5 |
|
Change the position from which BLOB content will be read, works only for stream BLOBs. |
|
2.0 |
interface is equivalent to the java.sql.Savepoint interface introduced in JDBC 3.0 specification, however allows using Firebird savepoints also in JDBC 2.0 (JDK 1.3.x) applications. |
|
|
|
|
|
Jaybird includes number of fixes that allow it pass JDBC 3.0 compatibility suite. It successfully passes 1216 tests, 60 tests were excluded, because they are either not applicable to Firebird or fail due to some server problems (math rounding issues, limitations of NUMERIC data type, etc.).
Jaybird driver has compile-time and run-time dependencies to JCA 1.0, JTA 1.0.1, JAAS 1.0 and JDBC 2.0 Optional Package. Additionally, if Log4J classes are found in the class path, it is possible to enable extensive logging inside the driver.
Following file groups can be found in distribution package:
Jaybird-2.1.0.jar
– archive containing JCA/JDBC driver and JMX management class.
It requires JCA 1.0, JTA 1.0.1, and JAAS 1.0.
Jaybird-pool-2.1.0.jar
– archive contains implementation of connection pooling and
statement pooling interfaces.
Jaybird-full-2.1.0.jar
– merge of firebirdsql.jar
and
mini-j2ee.jar
. This archive can be used
for standalone1
Jaybird deployments.
Jaybird-2.1.0.rar
– resource archive ready for deployment in JCA-enabled
application servers.
lib/jaas.jar
–
archive containing JAAS 1.0 classes.
lib/log4j-core.jar
– archive containing core Log4J classes that provide a
possibility to log into the file.
lib/mini-j2ee.jar
– archive containing JCA 1.0, JTA 1.0.1 and JDBC 2.0 Optional
Package classes.
jaybird21.dll
–
Windows version of the JNI library for Type 2 and Embedded Server
drivers.
libjaybird21.so
– Linux version of the JNI library for Type 2 and Embedded
Server drivers.
Jaybird JCA/JDBC driver is distributed under the GNU Lesser General Public License (LGPL). Text of the license can be obtained from http://www.gnu.org/copyleft/lesser.html. Using Jaybird (by importing Jaybird's public interfaces in your Java code), and extending Jaybird by subclassing or implemention of an extension interface (but not abstract or concrete class) is considered by the authors of Jaybird to be dynamic linking. Hence our interpretation of the LGPL is that the use of the unmodified Jaybird source does not affect the license of your application code.
Even more, all extension interfaces to which application might want to link are released under dual LGPL/modified BSD license. Latter is basically “AS IS” license that allows any kind of use of that source code. Jaybird should be viewed as an implementation of that interfaces and LGPL section for dynamic linking is applicable in this case.
Source code can be obtained from the CVS at
SourceForge.net. The CVSROOT is
:pserver:anonymous@cvs.sourceforge.net:/cvsroot/firebird
,
the module name is client-java
.
Alternatively source code can be viewed online at
http://cvs.sourceforge.net/viewcvs.py/firebird/client-java/
Driver provides different JDBC URLs for different usage scenarios:
jdbc:firebirdsql:host[/port]:/path/to/db.fdb
jdbc:firebirdsql://host[:port]/path/to/db.fdb
Default URL, will connect to the database using type 4 JDBC driver. Best suited for client-server applications with dedicated database server. Port can be omitted (default value is 3050), host name must be present.
First format is considered official, second – compatibility mode for InterClient migration.
jdbc:firebirdsql:native:host[/port]:/path/to/db.fdb
jdbc:firebirdsql:native://host[:port]/path/to/db.fdb
Type 2 driver, will connect to the database using
client library (either fbclient.dl
l or
gds32.dll
on Windows, and libfbclient.so
or libgds.so
on Linux). Requires correct
installation of the client library.
jdbc:firebirdsql:local:/path/to/db.fdb
Type 2 driver in local mode. Uses client library as in previous case, however will not use socket communication, but rather access database directly. Requires correct installation of the client library.
jdbc:firebirdsql:embedded:/path/to/db.fdb
Similar to the Firebird client library, however
fbembed.dll
on Windows and libfbembed.so
on Linux are used. Requires correctly installed and configured
Firebird embedded server.
Table below contains properties that specify
parameters of the connections that are obtained from this data
source. Commonly used parameters have the corresponding getter and
setter methods, rest of the Database Parameters Block parameters can
be set using setNonStandardProperty
setter method.
Property |
Getter |
Setter |
Description |
---|---|---|---|
|
+ |
+ |
Path to the database in the format
|
|
+ |
+ |
Type of the driver to use. Possible values are:
|
|
|
|
|
|
+ |
+ |
Size of the buffer used to transfer BLOB content. Maximum value is 64k-1. |
|
+ |
+ |
Size of the socket buffer. Needed on some Linux machines to fix performance degradation. |
|
+ |
+ |
Number of cache buffers (in database pages) that will be allocated for the connection. Makes sense for ClassicServer only. |
|
|
|
|
|
+ |
+ |
Character set for the connection. Similar to
|
|
+ |
+ |
Character encoding for the connection. See Firebird documentation for more information. |
|
+ |
+ |
Path to the properties file containg character translation map. |
|
|
|
|
|
+ |
+ |
Corresponding password. |
|
+ |
+ |
SQL role to use. |
|
+ |
+ |
Name of the user that will be used by default. |
|
|
|
|
|
+ |
+ |
Boolean flag tells driver whether stream BLOBs should be created by the driver, by default “false”. Stream BLOBs allow “seek” operation to be called, however due to a bug in gbak utility they are disabled by default. |
|
+ |
+ |
Boolean flag tells driver to assume that standard UDFs are defined in the database. This extends the set of functions available via escaped function calls. This does not affect non-escaped use of functions. |
|
+ |
+ |
Boolean flag tells driver to construct the default result set to be holdable. This prevents it from closing in auto-commit mode if another statement is executed over the same connection. |
|
|
|
|
|
+ |
+ |
TPB mapping for different transaction isolation modes. |
|
+ |
+ |
Default transaction isolation level. All newly created connections will have this isolation level. One of:
|
|
+ |
+ |
Integer value from |
|
|
|
|
|
|
+
|
Allows to set any valid connection property that does not have corresponding setter method. Two setters are available:
|
Jaybird 2.1 provides type 2 JDBC driver that uses native client library to connect to the databases. Additionally Jaybird 2.1 can use embedded version of Firebird relational database allowing to create Java applications that does not require separate server setup.
However type 2 driver has also limitations:
Due to multi-threading issues in Firebird client library as well as in embedded server version, it is not possible to access them from different threads simultaneously. When using client library only one thread is allowed to access connection at a time, however it is allowed to access different connections from different threads. Client library in local mode and embedded server library on Linux do not allow multithreaded access to the library. Jaybird provides necessary synchronization in Java code, however corresponding mutex is local to the classloader that loaded Jaybird driver. Care should be taken when deploying applications in web or application servers: put jar files in the main library directory of the web and/or application server, not in the library directory of the web or enterprise application (WEB-INF/lib directory or in the .EAR file). This issue will be fixed in Firebird client library after merge with Vulcan project.
Current implementation of the JNI library allows that either client library or embedded server library is loaded into memory. The first library being loaded wins, attempt to load second library will produce an error. Usually this is not a problem, since embedded server library can also act as a client library, so it is enough to load embedded server library first. This limitation will be removed in Jaybird 2.1, implementation already exists, however it was contributed already at RC stage; adding it to the release would delay it for few months.
Type 2 JDBC driver requires JNI library to be installed and available for Java Virtual Machine. Precompiled binaries for Windows and Linux platforms are distributed with Jaybird:
Please note that Jaybird 2.1 provides update to the JNI libraries to support new features. It is not compatible with JNI library for Jaybird 2.0.
jaybird21.dll
is
precompiled binary for Windows platform. Successfully tested with
Windows 2000 and Windows XP SP1, but there should be no issues also
in other Win32 OS.
Library should be either copied into the
directory specified in %PATH% environment variable, or made
available to JVM using the java.library.path
system property.
libjaybird21.so
is precompiled binary for Linux platform. It must be available via
the LD_LIBRARY_PATH environment variable, e.g. copied into /usr/lib/
directory. Another possibility is to specify path to the directory
with Jaybird JNI library in java.library.path
system property during the JVM startup.
Other platforms can easily compile the JNI
library by checking out the Jaybird sources from the CVS and using
“./build.sh compile-native
”
command in the directory with checked out sources.
After making Jaybird JNI library available to the
JVM application has to tell driver to start using this by either
specifying TYPE2 or LOCAL type in the connection pool or data source
properties or using appropriate JDBC URL when connecting via
java.sql.DriverManager
.
Embedded Server JDBC driver uses same JNI library and configuration steps for the type 2 JDBC driver.
There is however one issue related to the
algorithm of Firebird Embedded Server installation directory
resolution. Firebird server uses pluggable architecture for
internationalization. By default server loads fbintl.dll
or libfbintl.so
library that contains
various character encodings and collation orders. This library is
expected to be installed in the intl/
subdirectory of the server installation. The algorithm of directory
resolution is the following:
FIREBIRD
environment variable.
RootDirectory
parameter in the firebird.conf
file.
The directory where server binary is located.
When Embedded Server is used from Java and no
FIREBIRD
environment variable is
specified, it tries to find firebird.conf
in the directory where application binary is located. In our case
application binary is JVM and therefore Embedded Server tries to find
its configuration file in the bin/
directory of the JDK or JRE installation. Same happens to the last
item of the list. In most cases this is not desired behavior.
Therefore, if application uses
character encodings, UDFs or wants to fine-tune server's behavior
through the configuration file, the FIREBIRD
environment variable must be specified and point to the installation
directory of the Embedded Server, e.g. current working directory.
As it was mentioned before, Jaybird JCA/JDBC driver passed Sun JDBC CTS 1.3.1 test suite. All tests except those that do not apply to Firebird RDBMS succeeded. However driver is not officially JDBC-compliant because certification procedure is too expensive.
The following optional features and the methods for their support are not implemented:
java.sql.Array
data type is not yet supported.
java.sql.Blob
does not implement following methods:
position(Blob, long)
and position(byte[], long)
; Firebird
does not provide any server-side optimization for these calls,
client application must fetch complete BLOB content from the server
to do pattern search.
truncate(long)
;
Firebird does not provide such functionality on the server side,
application must fetch old BLOB from the server and pump old
content into a newly created BLOB.
java.sql.Clob
data type is not yet supported.
java.sql.Connection
getCatalog()
and setCatalog(String)
are not
supported by Firebird server.
getTypeMap()
and setTypeMap(Map)
are not supported.
prepareStatement(String,
int)
, prepareStatement(String, int[])
and prepareStatement(String, String[])
that return auto-generated keys are not implemented, because this
functionality is not provided by server.
java.sql.PreparedStatement
setObject(int index,
Object object, int type)
Target SQL type is determined from
the class of the passed object and corresponding parameter is
ignored.
setObject(int index,
Object object, int type, int scale)
Same as above, type and
scale are ignored.
java.sql.Ref
data
type is not supported by Firebird server.
java.sql.SQLData
data type is not supported by Firebird server.
java.sql.SQLInput
is not supported.
java.sql.SQLOutput
is not supported.
java.sql.Statement
cancel()
is not
supported by Firebird server.
execute(String, int)
,
execute(String, int[])
,
execute(String, String[])
,
executeUpdate(String, int)
,
executeUpdate(String, int[])
, and
executeUpdate(String, String[])
that
return auto-generated keys are not implemented, because of server
incapability.
getGeneratedKeys()
is not implemented because server does not support this feature
yet. As soon as support for this feature is available, driver will
be updated.
java.sql.Struct
data type is not supported by server.
The following methods are implemented, but deviate from the specification:
java.sql.Statement
get/setMaxFieldSize
does nothing, Firebird server does not support this feature.
get/setQueryTimeout
does nothing, Firebird server does not support this feature.
java.sql.PreparedStatement
setObject(int index,
Object object, int type)
Target SQL type is determined from
the class of the passed object and corresponding parameter is
ignored.
setObject(int index,
Object object, int type, int scale)
Same as above, type and
scale are ignored.
java.sql.ResultSetMetaData
isReadOnly()
always returns false
isWritable()
always returns true
isDefinitivelyWritable()
always returns true
Jaybird driver has also some implementation-specific issues that should be considered during development.
Jaybird behaves differently not only when different result set types are used but also the behavior depends whether connection is in auto-commit mode or not.
ResultSet.TYPE_FORWARD_ONLY
result sets when used in auto-commit mode are completely cached on
the client before the execution of the query is finished. This leads
to the increased time needed to execute statement, however the
result set navigation happens almost instantly. When auto-commit mode
is switched off, only part of the result set specified by the fetch
size is cached on the client.
ResultSet.TYPE_SCROLL_INSENSITIVE
result sets are always cached on the client. The reason is quite
simple – Firebird API does not provide scrollable cursor
support, navigation is possible only in one direction.
ResultSet.HOLD_CURSORS_OVER_COMMIT
holdability is supported in Jaybird only for result sets of type
ResultSet.TYPE_SCROLL_INSENSITIVE
. For
other result set types driver will throw an exception.
This interface can be used only to obtain
information about the IN parameters. Also it is not allowed to call
the PreparedStatement.getParameterMetaData
method before all of the OUT parameters are registered. Otherwise the
corresponding method of CallableStatement
throws an SQLException
, because the
driver tries to prepare the procedure call with incorrect number of
parameters.
Jaybird JDBC driver always uses connection encoding when converting array of bytes into character stream. The BLOB SUB_TYPE 1 fields allow setting the character encoding for the field. However when the contents of the field is sent to the client, it is not converted according to the character set translation rules in Firebird, but is sent “as is”. When such field is accessed from Java application via Jaybird and character set of the connection does not match the character encoding of the field, conversion errors might happen. Therefore it is recommended to convert such fields in the application using the appropriate encoding.
Current JCA implementation does not support
XAResource.forget(Xid)
method. It might
be important in cases when distributed transaction that was some time
in-limbo was either committed or rolled back by database
administrator. Such transactions appear to Jaybird as successfully
completed, however XA specification requires resource manager to
“remember” such transaction until the
XAResource.forget(Xid)
is called.
Connection pooling provides effective way to handle physical database connections. It is believed that establishing new connection to the database takes some noticeable amount or time and in order to speed things up one has to reuse connections as much as possible. While this is true for some software and for old versions of Firebird database engine, establishing connection is hardly noticeable with Firebird 1.0.3 and Firebird 1.5. So why is connection pooling needed?
There are few reasons for this. Each good connection pool provides a possibility to limit number of physical connections established with the database server. This is an effective measure to localize connection leaks. Any application cannot open more physical connections to the database than allowed by connection pool. Good pools also provide some hints where connection leak occurred. Another big advantage of connection pool is that it becomes a central place where connections are obtained, thus simplifying system configuration. However, main advantage of good connection pool comes from the fact that in addition to connection pooling, it can pool also prepared statement. Tests executed using AS3AP benchmark suite show that prepared statement pooling might increase speed of the application by 100% keeping source code clean and understandable.
When some statement is used more than one time, it makes sense to use prepared statement. It will be compiled by the server only once, but reused many times. It provides significant speedup when some statement is executed in a loop. But what if some prepared statement will be used during lifetime of some object? Should we prepare it in object's constructor and link object lifetime to JDBC connection lifetime or should we prepare statement each time it is needed? All such cases make handling of the prepared statements hard, they pollute application's code with irrelevant details.
Connection and statement pooling remove such details from application's code. How would the code in this case look like? Here's the example
Example 1. Typical JDBC code with statement pooling |
|
---|---|
001 002 003 004 005 006 007 008 009 010 011 012 013 014 015 016 |
... Connection connection = dataSource.getConnection(); try { PreparedStatement ps = connection.prepareStatement( “SELECT * FROM test_table WHERE id = ?”); try { ps.setInt(1, id); ResultSet rs = ps.executeQuery(); while (rs.next()) { // do something here } } finally { ps.close(); } } finally { connection.close(); } ... |
Lines 001-018 show typical code when
prepared statement pooling is used. Application obtains JDBC
connection from the data source (instance of javax.sql.DataSource
interface), prepares some SQL statement as if it is used for the
first time, sets parameters, and executes the query. Lines 012 and
015 ensure that statement and connection will be released under any
circumstances. Where do we benefit from the statement pooling? Call
to prepare a statement in lines 004-005 is intercepted by the pool,
which checks if there's a free prepared statement for the specified
SQL query. If no such statement is found it prepares a new one. In
line 013 prepared statement is not closed, but returned to the pool,
where it waits for the next call. Same happens to the connection
object that is returned to the pool in line 016.
Jaybird connection pooling classes belong to
org.firebirdsql.pool.*
package.
Description of some connection pool classes. |
|
---|---|
|
Base class for all connection pools. Can be used for implementing custom pools, not necessarily for JDBC connections. |
|
Subclass of |
|
Implementation of
|
|
Jaybird specific implementation of
|
|
Implementation of |
|
Implementation of |
|
Implementation of |
This class is a corner stone of connection and
statement pooling in Jaybird. It can be instantiated within the
application as well as it can be made accessible to other
applications via JNDI. Class implements both java.io.Serializable
and javax.naming.Referenceable
interfaces, which allows using it in a wide range of web and
application servers.
Class implements both
javax.sql.ConnectionPoolDataSource
and
javax.sql.XADataSource
interfaces.
Pooled connections returned by this class implement
javax.sql.PooledConnection
and
javax.sql.XAConnection
interfaces and
can participate in distributed JTA transactions.
Class provides following configuration properties:
This group contains properties defined in the JDBC specification and should be standard to all connection pools.
Property |
Getter |
Setter |
Description |
---|---|---|---|
|
+ |
+ |
Maximum time in milliseconds after which idle connection in the pool is closed. |
|
+ |
+ |
Maximum number of open physical connections. |
|
+ |
+ |
Minimum number of open physical connections. If value is greater than 0, corresponding number of connections will be opened when first connection is obtained. |
|
+ |
+ |
Maximum size of prepared statement pool. If 0, statement pooling is switched off. When application requests more statements than can be kept in pool, Jaybird will allow creating that statements, however closing them would not return them back to the pool, but rather immediately release the resources. |
This group of properties are specific to the Jaybird implementation of the connection pooling classes.
Property |
Getter |
Setter |
Description |
---|---|---|---|
|
+ |
+ |
Maximum time in milliseconds during which application can be blocked waiting for a connection from the pool. If no free connection can be obtained, exception is thrown. |
|
+ |
+ |
Period in which pool will try to obtain new connection while blocking the application. |
|
+ |
+ |
Allows to switch connection pooling off. |
|
+ |
+ |
Allows to switch statement pooling off. |
|
+ |
+ |
Statement that will be used to “ping” JDBC connection, in other words, to check if it is still alive. This statement must always succeed. |
|
+ |
+ |
Time during which connection is believed to be valid in any case. Pool “pings” connection before giving it to the application only if more than specified amount of time passed since last “ping”. |
This group contains read-only properties that provide information about the state of the pool.
Property |
Getter |
Setter |
Description |
---|---|---|---|
|
+ |
- |
Tells how many free connections are in the
pool. Value is between 0 and |
|
+ |
- |
Tells how many connections were taken from the pool and are currently used in the application. |
|
+ |
- |
Total size of open connection. At the pool
creation – 0, after obtaining first connection –
between |
|
+ |
- |
Deprecated. Same as |
This class is a wrapper for
FBConnectionPoolDataSource
converting
interface from javax.sql.ConnectionPoolDataSource
to javax.sql.DataSource
. It defines same
properties as FBConnectionPoolDataSource
does.
Pool implementation shipped with
Jaybird can provide hints for the application where the connection
was obtained from the pool, when it was released back to the pool,
when the statement was prepared. Such information is written into the
log when appropriate system properties are set to true
.
Additionally, when connection or prepared statement is closed twice,
driver will throw an SQL exception with an attached stack trace of
previous call to close()
method.
Property name |
Description |
---|---|
|
Enables logging inside driver. This is the
essential property, if it is not present or set to When it is set to
|
|
Enables logging of the thread stack trace when debugging is enabled and:
|
|
When statement caching is used and debugging is enabled, following information is logged:
|
The most detailed information can be found in the Jaybird Frequently Asked Questions (FAQ). The FAQ is included in the distribution, and is available on-line in several places.
Also a new resource, JaybirdWiki is available at http://Jaybirdwiki.firebirdsql.org.
The best place to start is the FAQ. Many details for using Jaybird with various programs are located there. Below are some links to useful web sites.
The http://groups.yahoo.com/group/Firebird-Java and corresponding mailing list Firebird-Java@yahoogroups.com.
The code for Firebird and this driver are on http://sourceforge.net/projects/firebird.
The Firebird project home page http://www.firebirdsql.com.
The developers follow the Firebird-Java@yahoogroups.com list. Join the list and post information about suspected bugs. This is a good idea because what is often thought to be a bug turns out to be something else. List members may be able o help out and get you going again, whereas bug fixes might take awhile.
If you are sure that this is a bug you may report it in the Firebird bug tracker, “Java Client (Jaybird)” at SourceForge.net project area (http://sourceforge.net/projects/firebird).
Please send corrections, suggestions, or additions to these Release Notes to to the mailing list at Firebird-Java@yahoogroups.com.
1You have to ensure that your class path contains JAAS 1.0 classes when using JDK 1.3.x.